Tjunction problem

[hucker]

I've had to set Tjunction to 80 to get the correct temperature reading.  This seems to have limited me to throttling at a maximum of 79C.  My CPU is happy at 90C, so I can't get the maximum out of it using Tthrottle.

Yes yes I know maybe it's not recommended, but I've always run i7s at 90C and they've never worn out.  They have autoshutoffs so they won't melt!

It's an Intel i7 940 on a Gigabyte X58A-UD3R motherboard.

[fred]

I've had to set Tjunction to 80 to get the correct temperature reading.  This seems to have limited me to throttling at a maximum of 79C.  My CPU is happy at 90C, so I can't get the maximum out of it using Tthrottle.

Yes yes I know maybe it's not recommended, but I've always run i7s at 90C and they've never worn out.  They have autoshutoffs so they won't melt!

It's an Intel i7 940 on a Gigabyte X58A-UD3R motherboard.

It's most likely 100C. Some older Cpu's have 80C.
Why do you think it's 80C ?
Check for this line in the logging. TJunction read from CPU: 100  °C, TJunction using: 100  °C
The first value is read from the CPU itself.

Try using V 5.43 beta.

90 C is high for a desktop not for a notebook.

[hucker]

Check for this line in the logging. TJunction read from CPU: 100  °C, TJunction using: 100  °C

TJunction read from CPU: 100  °C, TJunction using: 80  °C

It's most likely 100C. Some older Cpu's have 80C.
Why do you think it's 80C ?

I don't really understand what the 80 means that I'm entering, but setting it to 80 causes TThrottle to read the same temperature as my motherboard software.  I've confirmed this is right because the CPU will shut the power off at what the motherboard software says is 95C.

Try using V 5.43 beta.

Done, same problem.

90 C is high for a desktop not for a notebook.

I know, but I've never broken a CPU running at that temperature, 24/7/365.

[hucker]

Here's a screenshot to show you what I mean - TThrottle is not agreeing with my motherboard software on what temperature the CPU is running at.



It would appear (if I'm understanding the junction thingy correctly), that TThrottle is getting a reading from the tJunction output of the CPU which gives a value from say 0 to 1, 1 being "maximum" or 100C.  But when my motherboard software tells me that the CPU temperature is 90C, the tJunction output must be at about 110 or 1.1.  TThrottle does not accept anything over 1, so reads a lower temperature than it really is, and doesn't throttle when it should.

[fred]

It would appear (if I'm understanding the junction thingy correctly), that TThrottle is getting a reading from the tJunction output of the CPU which gives a value from say 0 to 1, 1 being "maximum" or 100C.  But when my motherboard software tells me that the CPU temperature is 90C, the tJunction output must be at about 110 or 1.1.  TThrottle does not accept anything over 1, so reads a lower temperature than it really is, and doesn't throttle when it should.

The junction is inside the package, so it's at least 10C higher that the reading on the package.

But do you see a line: TJunction read from CPU: in the TThrottle log tab?

[hucker]

Yes, see my post immediately above that one - I made 2.

"TJunction read from CPU: 100  °C, TJunction using: 80  °C"

(Although I then changed it to 100 before I did that screenshot to see what would happen).

So are you saying that the CPU actually cuts the power at 115C junction temperature?  And the motherboard software is just plain wrong?

[Checked with "speedfan"]
Speedfan agrees with TThrottle.  It seems the gigabyte motherboard software is reporting the OUTSIDE temperature of the CPU, how stupid of it!

So.... you won't let me throttle at 110C?  Oh go on.....

[Pepo]

It would appear (if I'm understanding the junction thingy correctly), that TThrottle is getting a reading from the tJunction output of the CPU which gives a value from say 0 to 1, 1 being "maximum" or 100C.  But when my motherboard software tells me that the CPU temperature is 90C, the tJunction output must be at about 110 or 1.1.  TThrottle does not accept anything over 1, so reads a lower temperature than it really is, and doesn't throttle when it should.

IIRC the core CPU temperature value is available as an integer (with 1 °C resolution), being a negative offset to the T_junc constant.

I personally doubt the offset may be negative (but I do not know for sure).

You can indeed  assume the BG SW does report the CPU package temperature, not the core chip temperature. It has for ages been like this (maybe also as an security precautions - the external sensor may be functional even if the CPU gets blocked). I have not followed the mainboards' development during last 5 years or so, to know, whether the MB software is already taking the core temp into account.

[hucker]

You can indeed assume the BG SW does report the CPU package temperature, not the core chip temperature.

I think what threw me was my old motherboard had poor software with it, so I used speedfan, which used Tjunction like Tthrottle does.

Interestingly, my old PC would lock up (presumably the CPU switched off) when the i7 920 reached 100C Tjunction.  This one cuts the power to the whole system at 115C Tjunction.  Mind you I have fiddled with the BIOS and switched off 2 of the three safety features :-)  For example i7s have their own throttling, but mine seems to throttle very early on, at a Tjunction of about 80C, so I switched that off.

Oh well, if I get this watercooler to work better I won't be running so hot anyway - I suspect the interface between the cpu  and the waterblock is not conducting heat that well - the waterblock doesn't feel that hot at all - mind you with room temperature water running through it that may be normal.  I've ordered some very high quality heatsink gunk.

[fred]

Normally the case temperature reading is a lot more accurate than the TJunction reading.
The Tj may even be dead wrong, so the case temperature is the benchmark.

But this all depends on overclocking, if you overclock the max case temperature goes down, as the inside Tj gets higher because of the OC.